144 righted-handed Korean elderly subjects with normal cognition were selected from participants of the Korean Longitudinal Study on Cognitive Aging and Dementia (KLOSCAD), which is a population-based prospective elderly cohort study on cognitive aging and dementia that was launched in 2009.³³ Participants of KLOSCAD were drawn from individuals aged 60 years or older. All participants assessed on clinical grounds to be cognitively normal in accordance with a Clinical Dementia Rating (CDR)³⁴ scale of 0. Participants who had Axis I disorders listed in the DSM-IV³⁵ or any serious medical or neurological disorders that could affect their cognitive function were excluded. Each subject was evaluated using the Korean version of the Consortium to Establish a Registry for Alzheimer's Disease (CERAD-K)³⁶ by a research geropsychiatrist with expertise in dementia research. The Edinburgh handedness inventory was used to assess handedness.³⁷ According to these criteria, we selected 144 righted-handed Korean elderly subjects with normal cognition for this study. This study protocol was approved by the Institutional Review Board of Seoul National University Bundang Hospital (SNUBH). All subjects were fully informed about the study protocol. All subjects provided written informed consents that were signed by the subjects.

The 144 subjects were divided into two groups. 96 subjects (M/F=48/48) aged 60 years or older (M=69.5±6.2 years, F=70.1±7.0 years) were used to construct a Korean normal elderly template (i.e., KNE96 template), and the remaining 48 subjects (M/F=24/24) aged 60 years of older (M=71.5±5.9 years, F=72.8±5.1 years) were used to validate the KNE96 template. The sex and age distribution of the two groups of subjects is shown in Table 1.

Three-dimensional (3D) T1-weighted spoiled gradient echo (SPGR) MRI images were acquired from all subjects on a 3.0Tesla GE SIGNA scanner (GE Medical Systems, Milwaukee, USA). The imaging parameters were as follows: 1.0 mm sagital slices, echo time (TE) of 3.68 ms, repetition time (TR) of 25.0 ms, flip angle of 30°, and field of view (FOV) of 24×24 cm. The acquisition matrix was 175×240×240 in the x-, y-, and z-dimensions yielding a voxel size of 1.0×1.0×1.0 mm³.

All data were acquired in DICOM format and then imported into ANALYZE format for analysis using the import function in SPM8. Each MRI scan was reoriented to virtual x-y-z coordinates so that the anterior commissural-posterior commissural (AC-PC) line of the brain was matched to the y-axis, the vertical anterior commissural (VAC) line to the z-axis, and the midsagittal plane to the y-z plane.

Global brain shape and size are useful baselines for characterizing and comparing individual brains, although they do not provide detailed morphometry. Length, width, height, and volume of the whole brain are used as representative features of global brain shape and size. We measured the global brain features of each brain of 96 subjects in the construction group. We manually specified six reference points using BrainSuite:³⁸³⁹ the most anterior (AP), posterior (PP), right (RP), left (LP), superior (SP), and inferior (IP) points of the cerebrum. The length, width, and height of the whole brain were determined as the distances between AP and PP, RP and LP, and SP and IP, respectively. Following this, the width to length (W/L), height to length (H/L), and height to width (H/W) ratios were computed. We automatically extracted tissue probability maps of white matter (WM), grey matter (GM), and cerebrospinal fluid (CSF) using the segmentation routine in SPM8.⁴⁰ Whole brain volume was determined as the number of 1.0 mm³ isovoxels within which the sum of the three tissue probability maps was over 0.5.⁴¹⁴²

We constructed the KNE96 template, composed of high resolution 3D structural MR images from 96 Korean normal elderly subjects, using the DARTEL toolbox in SPM8.⁴³ First, each of MRI scans was segmented using the segmentation routine in SPM8, and then the spatial transformation files derived from the segmentation routine was used to import the MRI scan into the rigidly transformed tissue probability maps of WM, GM and CSF. Following this, the tissue types were simultaneously registered (i.e., WM with WM, GM with GM, and CSF with CSF). The initial template is constructed as the mean of these 96 MRI images in the construction group. Deformations from this template to each of the individual images were computed and then the template was updated by applying the inverse deformations to the images and averaging the deformed images. This procedure was repeated based on multi-level pyramid approach, beginning with a more coarse registration then ending with the more detailed registration. After a total of 36 iterations, three probability maps of the KNE96 template were finally constructed. Finally, for T1 image of the KNE96 template, we warped individual T1 images using the flow fields that store deformation information and created the mean of the 96 warped T1 images.

In addition, we also constructed two gender-specific templates (KNE48M template for male and KNE48F template for female) by dividing the subjects in the construction group into two subgroups by gender and repeating the template construction process for 48 male subjects and 48 female subjects, respectively.

To validate the KNE96 template, we used a nonlinear version of the ICBM152 atlas (i.e., ICBM152 template) as a comparison template.¹⁵¹⁶ We compared the deformations that resulted from image registration using the KNE96 and ICBM 152 templates. Each MRI scan in the validation group was registered to both the KNE96 and ICBM152 templates using DARTEL.⁴³ We used displacement vectors and Jacobian determinants to analyze the deformation of each subject to each template.⁴⁴ For each voxel, the Euclidean distance between the subject and its deformed image was obtained as the magnitude of the displacement vector, dv→. Furthermore, the Jacobian determinant is an indicator of volumetric change at each voxel under deformation. Its value is >1, 1, or <1 if there is volume expansion, preservation, or contraction, respectively. Therefore, we derived two kinds of maps for each test subject registered to each template: the magnitude of displacement vectors (i.e., mag-displacement) and log of Jacobian determinants (i.e., log-Jacobian).

We analyzed the differences in the mag-displacement and log-Jacobian maps between the KNE96 and ICBM152 templates using voxel-based morphometry (VBM). Because the mag-displacement and log-Jacobian maps were generated in each template space that was used as the reference for registration, these maps were spatially normalized so that the coordinates were in MNI space, using the DARTEL toolbox.⁴³ Comparison between the KNE96- and ICBM152-registered mag-displacement and log-Jacobian maps was performed with a paired-samples t-test for each voxel. With Bonferroni correction, p<6.9×10^-9 was considered a statistically significant difference.

In addition to VBM analysis of the mag-displacement and log-Jacobian maps, to explore the region-specific differences between the KNE96 and ICBM152 templates, we determined the deformation for each anatomical region of the Talairach atlas.¹¹⁴⁵ First, voxels were classified into 1105 anatomical regions using the Talairach atlas. Second, for each Talairach region, the significant averaged differences (SADs) of the mag-displacement and the log-Jacobian were computed as follows: with:

Here, x_i represents the ith voxel in the mag-displacement and log-Jacobian maps and N_Rt represents the number of voxels that belong to the tth Talairach region, R_t. S_j and N_s=48 represent the jth subject and the number of subjects in the validation group, respectively. Significant voxel rates, R_{mag-displacemen} (R_t) and R_log-jacobian (R_t), were defined as the ratio of the number of significant voxels from the paired-samples t-tests to the total number of voxels in the Talairach region, R_t.

Figure 1 shows the KNE96, KNE48M, KNE48F, and IC BM152 templates. In Korean subjects, male brains were longer, wider, taller, and thus larger than the female brains. However, the ratios of W/L, H/L, and H/W were comparable between male and female brains (Table 2). The KNE96 template was shorter but wider than the ICBM152 template (Table 3), which was consistent with previous observations that report differences in global hemispheric shape and size between Occidental and Oriental brains.²⁹³⁰⁴⁶

Figure 2 shows the differences in the mag-displacement and log-Jacobian maps between the KNE96 and ICBM152 templates. As shown in Figure 2A, we found that the mag-displacement and log-Jacobian induced by registration to the KNE96 template were much smaller than those induced by registration to the ICBM152 template in most brain regions. Figure 2B illustrates the significant voxels of the VBM analysis overlaid on corresponding standard T1 image. This indicated that there is less deformation and a lower risk of volumetric errors if KNE96 is used as the reference template in neuroimaging studies in the Korean elderly population.

Figure 3 shows the SAD maps of the mag-displacement and the log-Jacobian. The mag-displacement SAD map shows large values in some Talairach regions within the frontal and occipital lobes while the log-Jacobian SAD map shows large values in the frontal and occipital lobes, and the thalamus.

By considering each Talairach region as a data point, we constructed a two-dimensional scatter plot in which the y-axis is the mag-displacement SAD and the x-axis is the log-Jacobian SAD, as shown in Figure 4. Because we excluded small Talairach regions that had a volume less than 500 mm³, Figure 4 contains 355 data points. And then, we performed a two-step cluster analysis in PASW Statistics 18.0 (IBM SPSS Inc, 2009, Chicago, IL, USA) to partition these data points relative to their mag-displacement and log-Jacobian SAD values. Tables 4 and 5 list the Talairach regions with large mag-displacement and log-Jacobian SADs, respectively. Volume cha-nge indicates the ratio of the deformed volume using the ICBM152 template to that using the KNE96 template. The Talairach regions with large mag-displacement SAD values (circles in Figure 4) include the superior, medial, and middle frontal gyrus in the frontal lobe, lingual gyrus and cuneus in the occipital lobe, and caudate in the sub-lobar (Table 4). The Talairach regions with large log-Jacobian SAD values (boxes in Figure 4) include the thalamus in the sub-lobar, superior, middle, and inferior occipital gyrus in the occipital lobe, rectal gyrus in the frontal lobe, inferior semilunar lobule in the posterior lobe, and fusiform gyrus in the temporal lobe (Table 5). These regions may be subject to considerably larger changes in position or volume if the ICBM152 was used instead of the KNE96 as a reference template for neuroimaging studies in elderly Koreans.